On Thu, Jun 02, 2016 at 10:25:46PM +0200, Daniel Vetter wrote:
On Thu, Jun 02, 2016 at 07:28:30PM +0200, Philipp Zabel wrote:
Am Donnerstag, den 02.06.2016, 18:21 +0200 schrieb Daniel Vetter:
[...]
Which means when we drop that 1 reference in the saved connector (which is entirely bogus code, I agree) it should only drop to 1, not 0. And youre cleanup code should not be called.
The other thing is that radeon/nouveau also uses this code, and no one complained yet. Together I think that's some good evidence that suggests there's also something strange going on in imx itself.
That's what I wonder, too. Right now I can't see it though.
Yep, something funny is definitely going on here.
Is this perhaps something similar to what Jon fixed a couple of weeks ago in the Tegra driver? See commit 280dc0e145b2 ("drm/tegra: Fix crash caused by reference count imbalance").
The root cause of that was that Tegra's DSI driver is subclassing the connector state and didn't make proper use of the helpers to share the core state code, ultimately causing the connector reference count to become unbalanced. dw-hdmi doesn't subclass state, so it seems unlikely that it's the same root cause here. The kernel version string indicates that this isn't a vanilla v4.7-rc1 but that you have local changes, so perhaps there's some significant change there?
Thierry